xref: /linux/net/ipv4/netfilter.c (revision cc04a46f11ea046ed53e2c832ae29e4790f7e35f)
1 /*
2  * IPv4 specific functions of netfilter core
3  *
4  * Rusty Russell (C) 2000 -- This code is GPL.
5  * Patrick McHardy (C) 2006-2012
6  */
7 #include <linux/kernel.h>
8 #include <linux/netfilter.h>
9 #include <linux/netfilter_ipv4.h>
10 #include <linux/ip.h>
11 #include <linux/skbuff.h>
12 #include <linux/gfp.h>
13 #include <linux/export.h>
14 #include <net/route.h>
15 #include <net/xfrm.h>
16 #include <net/ip.h>
17 #include <net/netfilter/nf_queue.h>
18 
19 /* route_me_harder function, used by iptable_nat, iptable_mangle + ip_queue */
20 int ip_route_me_harder(struct sk_buff *skb, unsigned int addr_type)
21 {
22 	struct net *net = dev_net(skb_dst(skb)->dev);
23 	const struct iphdr *iph = ip_hdr(skb);
24 	struct rtable *rt;
25 	struct flowi4 fl4 = {};
26 	__be32 saddr = iph->saddr;
27 	__u8 flags = skb->sk ? inet_sk_flowi_flags(skb->sk) : 0;
28 	unsigned int hh_len;
29 
30 	if (addr_type == RTN_UNSPEC)
31 		addr_type = inet_addr_type(net, saddr);
32 	if (addr_type == RTN_LOCAL || addr_type == RTN_UNICAST)
33 		flags |= FLOWI_FLAG_ANYSRC;
34 	else
35 		saddr = 0;
36 
37 	/* some non-standard hacks like ipt_REJECT.c:send_reset() can cause
38 	 * packets with foreign saddr to appear on the NF_INET_LOCAL_OUT hook.
39 	 */
40 	fl4.daddr = iph->daddr;
41 	fl4.saddr = saddr;
42 	fl4.flowi4_tos = RT_TOS(iph->tos);
43 	fl4.flowi4_oif = skb->sk ? skb->sk->sk_bound_dev_if : 0;
44 	fl4.flowi4_mark = skb->mark;
45 	fl4.flowi4_flags = flags;
46 	rt = ip_route_output_key(net, &fl4);
47 	if (IS_ERR(rt))
48 		return PTR_ERR(rt);
49 
50 	/* Drop old route. */
51 	skb_dst_drop(skb);
52 	skb_dst_set(skb, &rt->dst);
53 
54 	if (skb_dst(skb)->error)
55 		return skb_dst(skb)->error;
56 
57 #ifdef CONFIG_XFRM
58 	if (!(IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED) &&
59 	    xfrm_decode_session(skb, flowi4_to_flowi(&fl4), AF_INET) == 0) {
60 		struct dst_entry *dst = skb_dst(skb);
61 		skb_dst_set(skb, NULL);
62 		dst = xfrm_lookup(net, dst, flowi4_to_flowi(&fl4), skb->sk, 0);
63 		if (IS_ERR(dst))
64 			return PTR_ERR(dst);
65 		skb_dst_set(skb, dst);
66 	}
67 #endif
68 
69 	/* Change in oif may mean change in hh_len. */
70 	hh_len = skb_dst(skb)->dev->hard_header_len;
71 	if (skb_headroom(skb) < hh_len &&
72 	    pskb_expand_head(skb, HH_DATA_ALIGN(hh_len - skb_headroom(skb)),
73 				0, GFP_ATOMIC))
74 		return -ENOMEM;
75 
76 	return 0;
77 }
78 EXPORT_SYMBOL(ip_route_me_harder);
79 
80 /*
81  * Extra routing may needed on local out, as the QUEUE target never
82  * returns control to the table.
83  */
84 
85 struct ip_rt_info {
86 	__be32 daddr;
87 	__be32 saddr;
88 	u_int8_t tos;
89 	u_int32_t mark;
90 };
91 
92 static void nf_ip_saveroute(const struct sk_buff *skb,
93 			    struct nf_queue_entry *entry)
94 {
95 	struct ip_rt_info *rt_info = nf_queue_entry_reroute(entry);
96 
97 	if (entry->state.hook == NF_INET_LOCAL_OUT) {
98 		const struct iphdr *iph = ip_hdr(skb);
99 
100 		rt_info->tos = iph->tos;
101 		rt_info->daddr = iph->daddr;
102 		rt_info->saddr = iph->saddr;
103 		rt_info->mark = skb->mark;
104 	}
105 }
106 
107 static int nf_ip_reroute(struct sk_buff *skb,
108 			 const struct nf_queue_entry *entry)
109 {
110 	const struct ip_rt_info *rt_info = nf_queue_entry_reroute(entry);
111 
112 	if (entry->state.hook == NF_INET_LOCAL_OUT) {
113 		const struct iphdr *iph = ip_hdr(skb);
114 
115 		if (!(iph->tos == rt_info->tos &&
116 		      skb->mark == rt_info->mark &&
117 		      iph->daddr == rt_info->daddr &&
118 		      iph->saddr == rt_info->saddr))
119 			return ip_route_me_harder(skb, RTN_UNSPEC);
120 	}
121 	return 0;
122 }
123 
124 __sum16 nf_ip_checksum(struct sk_buff *skb, unsigned int hook,
125 			    unsigned int dataoff, u_int8_t protocol)
126 {
127 	const struct iphdr *iph = ip_hdr(skb);
128 	__sum16 csum = 0;
129 
130 	switch (skb->ip_summed) {
131 	case CHECKSUM_COMPLETE:
132 		if (hook != NF_INET_PRE_ROUTING && hook != NF_INET_LOCAL_IN)
133 			break;
134 		if ((protocol == 0 && !csum_fold(skb->csum)) ||
135 		    !csum_tcpudp_magic(iph->saddr, iph->daddr,
136 				       skb->len - dataoff, protocol,
137 				       skb->csum)) {
138 			skb->ip_summed = CHECKSUM_UNNECESSARY;
139 			break;
140 		}
141 		/* fall through */
142 	case CHECKSUM_NONE:
143 		if (protocol == 0)
144 			skb->csum = 0;
145 		else
146 			skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
147 						       skb->len - dataoff,
148 						       protocol, 0);
149 		csum = __skb_checksum_complete(skb);
150 	}
151 	return csum;
152 }
153 EXPORT_SYMBOL(nf_ip_checksum);
154 
155 static __sum16 nf_ip_checksum_partial(struct sk_buff *skb, unsigned int hook,
156 				      unsigned int dataoff, unsigned int len,
157 				      u_int8_t protocol)
158 {
159 	const struct iphdr *iph = ip_hdr(skb);
160 	__sum16 csum = 0;
161 
162 	switch (skb->ip_summed) {
163 	case CHECKSUM_COMPLETE:
164 		if (len == skb->len - dataoff)
165 			return nf_ip_checksum(skb, hook, dataoff, protocol);
166 		/* fall through */
167 	case CHECKSUM_NONE:
168 		skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, protocol,
169 					       skb->len - dataoff, 0);
170 		skb->ip_summed = CHECKSUM_NONE;
171 		return __skb_checksum_complete_head(skb, dataoff + len);
172 	}
173 	return csum;
174 }
175 
176 static int nf_ip_route(struct net *net, struct dst_entry **dst,
177 		       struct flowi *fl, bool strict __always_unused)
178 {
179 	struct rtable *rt = ip_route_output_key(net, &fl->u.ip4);
180 	if (IS_ERR(rt))
181 		return PTR_ERR(rt);
182 	*dst = &rt->dst;
183 	return 0;
184 }
185 
186 static const struct nf_afinfo nf_ip_afinfo = {
187 	.family			= AF_INET,
188 	.checksum		= nf_ip_checksum,
189 	.checksum_partial	= nf_ip_checksum_partial,
190 	.route			= nf_ip_route,
191 	.saveroute		= nf_ip_saveroute,
192 	.reroute		= nf_ip_reroute,
193 	.route_key_size		= sizeof(struct ip_rt_info),
194 };
195 
196 static int __init ipv4_netfilter_init(void)
197 {
198 	return nf_register_afinfo(&nf_ip_afinfo);
199 }
200 subsys_initcall(ipv4_netfilter_init);
201